1 Data

The image data was read in as two 3-D matrices. The first one if with intensity value, the second one a binary flag, with one indicating a lesion voxel and zero a normal voxel. Please not that the lesion flag was obtained through a automatic segmentation process. As a consequence, there could be false positive flags.

Below is an example of one subject (01-001). The red part are voxels labeled as lesion.

2 Preliminary analysis

The current plan is to study the correlation between scans done at each specific area. A separate linear regression model of scan 2 on scan 1 is fit on each subject at each site, and only voxels that are labeled as “positive” in both scans are used in the regression.

Below is a detailed procedure for regression on each subject at each site:

  1. pick out voxels that are labeled positive in both scan.
  2. standardize the intensity value
  3. fit regression a model calculate pearson correlation between scans from the same subject at the same site

Below is a display of the results. If any subject did not received a scan at any site, the corresponding figure is left empty.

The major concern with this output is the lack of correlation across scans from the same subject at the same site. As the figure shows, the correlation value is very close two zero for most of these figures.